[FFmpeg-devel] [libav-devel] [PATCH] h264: update avctx width/height when flushing

[Vittorio Giovara]

On Fri, Jun 12, 2015 at 9:01 PM, Andreas Cadhalpun
<andreas.cadhalpun at googlemail.com> wrote:
> On 12.06.2015 21:51, Hendrik Leppkes wrote:
>> On Fri, Jun 12, 2015 at 9:23 PM, Andreas Cadhalpun
>>> I agree that it is not really elegant, but removing the assumption that
>>> the avctx width/height/pix_fmt are from the last decoded frame instead
>>> of the last returned frame from the h264 decoder is beyond me.
>>> If someone more familiar with the h264 decoder could do that, it would
>>> be great.
>>> But until then, having this workaround is better than the current state.
>>> So I've pushed this.
>>>
>>
>> This patch is super ugly and a mega hack.
>
> Maybe. ;)
>
>> Just inform consumers that only AVFrame has the actual up-to-date
>> values when a change in the bitstream happens and stop doing such
>> crap.
>
> That's not currently documented, so it would effectively be an API change.
> And since the h264 decoder is the only one misbehaving in this way,
> fixing it is the better alternative.

Not necessarily, a lot of parameters are valid only in the avframe
while in the context they might not be completely updated.
The first that comes to mind is interlaced_frame -- there are 3 places
this might (or might not) be signalled (context, frame and stream or
parser iirc) but the RightWay is to check it from the frame only.

Do you have a test case where using the API exposes some kind of problem?
-- 
Vittorio